Keith Joung remembers the first time he took CRISPR for a spin. In late 2012, the pathologist at Massachusetts General Hospital in Boston assembled the components of the new gene-editing technology and fiddled with the DNA of a zebrafish embryo. “It was so easy to do,” he says. “It was just stunning.”
CRISPR—the highly efficient set of molecular scissors recently selected as Science’s Breakthrough of the Year—might be easy to use, but it’s not perfect. Joung and his colleagues soon found that these scissors could get too slice-happy, cutting DNA in unexpected and unwanted locations. In early experiments, the group observed that these off-target effects could occur at some DNA sites with nearly the same frequency as the desired edits. That’s a problem if CRISPR is to form the basis of human therapies, for example, repairing the defective genes that cause muscular dystrophy or hereditary liver disease. Researchers’ primary concern is that cutting into an unwanted gene could cause uncontrolled growth and cancer.
Now, Joung and colleagues have found a way to make CRISPR more precise. In a new study, they modified its cutting enzyme to reduce off-target effects below detectable levels.
“I think that this is a potential breakthrough,” says Jin-Soo Kim, a molecular biologist at Seoul National University who was not involved with the work. But the quest to perfect CRISPR doesn’t have a clear end. “No drugs are free of off-target effects,” he notes. With CRISPR-based therapies still far from human testing, no one knows just how precise is precise enough.
CRISPR relies on a DNA-cutting enzyme called Cas9 attached to a short strand of RNA that guides it to specific point in the genome. When the RNA finds a complementary—or nearly complementary—sequence, Cas9 makes its slice. There are already several approaches to prevent unintended slicing. Shortening the length of the guide RNA makes it more sensitive to mismatched sequences, but it can also create entirely new off-target effects. Some labs have experimented with a version of Cas9 that cuts through a single DNA strand instead of two. That means two Cas9 enzymes bearing two different guide RNAs have to recognize their target sequences to cut both strands—a more demanding matching process. But doubling the number of RNA guides adds bulk, which could make it harder to deliver a CRISPR-based treatment into cells.
In the new work, published online today in Nature, Joung and colleagues took a different approach. They modified the Cas9 enzyme itself to change the way it interacts with DNA. They first altered some of the “residues” on the enzyme’s surface that presumably help the guide RNA pair with its matching DNA strand. One set of modifications created a new variant of Cas9, called Cas9-HF1, that appears to be much more discriminating in its cuts. The researchers made seven different edits guided by seven different RNA strands, each known to produce off-target effects with Cas9. But Cas9-HF1 showed no detectable off-target effects in six of these cases—and just one errant slice in the seventh, they report. Joung adds that the apparent slice could actually be the result of a sequencing error.
The results come on the heels of a similar feat, led by CRISPR pioneer Feng Zhang of Harvard University and the Broad Institute in Cambridge, Massachusetts, published last month in Science. That team modified Cas9 to change how it interacts with a different part of a cell’s DNA. It, too, dramatically improved CRISPR’s specificity. But it’s hard to compare those results directly with the new paper because they used slightly different methods to measure off-target effects.
Joung claims his group’s measurements are roughly 10-fold more sensitive than the one used in the Science paper. Both studies rely on methods that attach molecular tags to all points in the genome where a double-stranded break has occurred, before sequencing the short, flagged segments to count the cuts in various genes. Joung’s team claims to detect edits that occur in at least 0.1% of the genome. Zhang says the method used in his paper has been validated down 0.3%, and it may be even more sensitive.
Does detecting just a couple of faulty cuts in a thousand matter? Absolutely, Joung says. “A lot of therapeutic strategies envision manipulating millions, tens of millions, even hundreds of millions of cells, potentially. So one in 1000 sounds pretty good, but that number can become quite large.” He argues that the field needs tests that root out these potentially harmful effects at frequencies of 0.01% or even lower.
Others are less focused on increasingly sensitive tests. Because CRISPR will never fully be rid of off-target effects, the key question for a given therapy is not strictly how many unwanted cuts it makes, but whether it disrupts any essential genes, says Jiing-Kuan Yee, a molecular biologist at the research center City of Hope in Duarte, California. Each therapeutic application will require its own carefully selected Cas9 molecule—and modifications like those in the two recent papers might be combined.
“Pretty soon, I think everybody’s going to start using these modified Cas9s,” he says. “The [off-target] problem will still be there, but it’s going to be much, much reduced.”